| { |
| "name": "SciFIBench", |
| "release_date": "2024-09-01", |
| "subsets": { |
| "figure2caption": { |
| "language": [ |
| "en" |
| ], |
| "modalities": [ |
| "single_image_start" |
| ], |
| "task_type": "multiple_choice_qa", |
| "score_pipeline": [ |
| "rule-match", |
| "llm-match" |
| ], |
| "score_protocol": { |
| "reference": "official@github.com/jonathan-roberts1/SciFIBench — extract_answer() uses an LLM (Gemini-Pro 1.0, temperature 0, max 3 tokens) with prompt 'Return only the letter of the answer' to parse the chosen option A-E from model output, then accuracy = extracted letter == gold letter.", |
| "note": "Official extractor model is Gemini-Pro 1.0 (LLM used for extraction only, compare is exact letter match); our chain prepends a deterministic prefetch, a faithful superset." |
| }, |
| "prompt_template": "<image>{{ question }}\n{% for opt in options %}{{ opt }}\n{% endfor %}Answer with the option letter (A, B, C, D, or E).", |
| "mapping_from_source": { |
| "source": { |
| "format": "huggingface", |
| "url": { |
| "CS_Figure2Caption": "https://huggingface.co/datasets/jonathan-roberts1/SciFIBench", |
| "General_Figure2Caption": "https://huggingface.co/datasets/jonathan-roberts1/SciFIBench" |
| } |
| }, |
| "id": { |
| "from": "ID" |
| }, |
| "question": { |
| "from": "Question" |
| }, |
| "options": { |
| "from": "Options" |
| }, |
| "answer": { |
| "from": "Answer" |
| }, |
| "media": { |
| "from": "Images", |
| "type": "list", |
| "min_items": 1, |
| "max_items": 1 |
| }, |
| "extra": { |
| "category": { |
| "from": "Category" |
| }, |
| "source_id": { |
| "from": "source_id", |
| "optional": true, |
| "note": "source column present on a sibling split; empty string here — column added for cross-split Arrow schema uniformity (push_to_hub requires identical features per config)" |
| } |
| } |
| }, |
| "prompt_template_source": { |
| "origin": "official", |
| "reference": "https://github.com/jonathan-roberts1/SciFIBench (README example prompts: question + lettered options + 'Answer with the option letter (A, B, C, D, or E).')", |
| "notes": "Tier 1: SciFIBench README example prompts as published by the dataset authors." |
| } |
| }, |
| "caption2figure": { |
| "language": [ |
| "en" |
| ], |
| "modalities": [ |
| "multi_image_start" |
| ], |
| "task_type": "multiple_choice_qa", |
| "score_pipeline": [ |
| "rule-match", |
| "llm-match" |
| ], |
| "score_protocol": { |
| "reference": "official@github.com/jonathan-roberts1/SciFIBench — same protocol as figure2caption: LLM (Gemini-Pro) letter extraction then exact-letter accuracy; caption2figure options are 5 candidate figures labeled A-E (5 images per row).", |
| "note": "Same as figure2caption; official extractor is Gemini-Pro 1.0." |
| }, |
| "prompt_template": "<image><image><image><image><image>{{ question }}\n{% for opt in options %}{{ opt }}\n{% endfor %}Answer with the option letter (A, B, C, D, or E).", |
| "mapping_from_source": { |
| "source": { |
| "format": "huggingface", |
| "url": { |
| "CS_Caption2Figure": "https://huggingface.co/datasets/jonathan-roberts1/SciFIBench", |
| "General_Caption2Figure": "https://huggingface.co/datasets/jonathan-roberts1/SciFIBench" |
| } |
| }, |
| "id": { |
| "from": "ID" |
| }, |
| "question": { |
| "from": "Question" |
| }, |
| "options": { |
| "from": "Options" |
| }, |
| "answer": { |
| "from": "Answer" |
| }, |
| "media": { |
| "from": "Images", |
| "type": "list", |
| "min_items": 5, |
| "max_items": 5 |
| }, |
| "extra": { |
| "category": { |
| "from": "Category" |
| }, |
| "source_id": { |
| "from": "source_id", |
| "optional": true, |
| "note": "source column present on a sibling split; empty string here — column added for cross-split Arrow schema uniformity (push_to_hub requires identical features per config)" |
| } |
| } |
| }, |
| "prompt_template_source": { |
| "origin": "official", |
| "reference": "https://github.com/jonathan-roberts1/SciFIBench (README example prompts: question + lettered options + 'Answer with the option letter (A, B, C, D, or E).')", |
| "notes": "Tier 1: SciFIBench README example prompts as published by the dataset authors." |
| } |
| } |
| } |
| } |
